The publication shows that it is not a simple colour correction: http://openaccess.thecvf.com/content_CVPR_2019/papers/Akkayn...
Example: https://imgur.com/a/h3uFu4z
That simple, huh?
I don't understand why to even say something like that. PR value? As I understood the paper, the presented method is very refreshingly clean of anything resembling what we call AI today. It seems to be a combination of good, old-school photogrammetry and image processing techniques - which is great, because with such methods they can actually ensure the result is physically correct.
[0] http://openaccess.thecvf.com/content_CVPR_2019/papers/Akkayn...
also, since the technique is removing a foggy haze, it seems like this could be used for selfdriving cars, with multiple cameras along the periphery of the car, to clean the image for foggy conditions (fog, smoke, smog, ...)
It's sort of like how having a big image corpus enabled AI. There's no AI in the images themselves, but it gives you something to throw AI at.
Yes, the hype around AI / machine learning is such that everyone wants it to do magic and invigorate their field of study.
sure it's more aesthetically pleasing, but it's already distorting reality
personally I think photos should be most accurate representation of what healthy naked human eye see, no beatification, no bigger contrast or oversaturating despite making photos more appealing. if you twist reality where it will stop, where is the border of what is too much?
This is not what the eye sees, but in general, very, very few photos are. Do you also model the fovea in photos you take and make everything blurry around the center in the pursuit of greater realism? Do you shun long exposures of the night sky, as our eye can't see the milky way in that detail anyway?
It's a misguided goal in general, I'd say. Some photos aim to realistically depict a scene, some aim to artistically depict a scene, some aim to depict a scene to evoke certain emotions and either realistic or artistic might be valid choices for that. The list can go on, I guess.
In this case, being able to see and convey the actual colors of what is being photographed has scientific and artistic value.
Under what lighting?
An alternative can be to make some similar looking objects that are like fake coral blocks and use a boat to sumerge them in the sea and compare the corrected photos with the photos taken in air before. (The wet surfaces have a different look, and it would be important to use a diffuse light instead of direct sunlight, and perhaps other technical problems for the comparison.)
Not with this technique at least. Maybe you could train a CNN to do that...
I've done this in Darkroom and Blender (with the node compositor).
In particular, a fog machine will help emulate the haze and backscatter produced by water and the particles floating in it.
As a primarily video guy, I always laugh to myself (sometimes not to myself) at the amount of effort photo editors spend on a single image. I remind them that the video world has to do that same level of work, except x24 per second multiplied by number of seconds. Photoshop is cool, but Nuke is mindblowing
The two projects seems unrelated at first glance, but the timing is interesting.
Original article linked in current article, has more pictures/content.
The one thing that is disturbing though is images with visible "horizon" (for lack of a better word). I find the images that look like they were taken on land but don't have a sky where one would expect it somewhat uncanny.
